Scatter/Gather Clustering: Flexibly Incorporating User Feedback to Steer Clustering Results

Area of Science:

• Data Science
• Computational Biology
• Human-Computer Interaction

Background:

• Traditional clustering algorithms often lack user expressiveness.
• Integrating domain knowledge into clustering is challenging.
• Existing methods struggle with dynamic restructuring of cluster results.

Purpose of the Study:

• To introduce scatter/gather clustering, an interactive method for users to refine clustering outcomes.
• To develop a framework that supports expressive user interactions for data restructuring.
• To enable effective incorporation of domain expertise into clustering processes.

Main Methods:

• Developed a nonlinear optimization framework for scatter/gather clustering.
• Implemented dual primitives: 'scatter' (break apart clusters) and 'gather' (merge clusters).
• Combined scatter and gather operations for dynamic data restructuring.

Main Results:

• Demonstrated scatter/gather clustering's effectiveness in a visual analytic application for bat biosonar baffle shape analysis.
• Showcased how domain experts can supply intuitive scatter/gather constraints.
• Validated the framework's ability to incorporate constraints without numerous instance-level requirements.

Conclusions:

• Scatter/gather clustering offers a powerful and intuitive way for users to guide and refine data clustering.
• The nonlinear optimization framework effectively balances cluster locality with user-defined constraints.
• This approach significantly enhances the utility of clustering for domain-specific scientific applications.